Description
Plain RAG retrieves once and answers, so vague questions get off-target replies. This project builds a modular Agentic RAG system with LangGraph, with conversation memory and human-in-the-loop clarification before retrieving.
It combines vector search and BM25 on Qdrant, supports Ollama and several providers, and ships a Gradio UI you can run and understand in minutes.
Agentic retrieval:Multi-step search and reasoning.
Clarification:Confirms intent with the user.
Hybrid search:Vectors plus BM25.
Quick start:Gradio UI and local models.
It combines vector search and BM25 on Qdrant, supports Ollama and several providers, and ships a Gradio UI you can run and understand in minutes.
Features
Agentic retrieval:Multi-step search and reasoning.
Clarification:Confirms intent with the user.
Hybrid search:Vectors plus BM25.
Quick start:Gradio UI and local models.
